This paper introduces “failure before violation” as a distinct failure mode in constrained systems. Traditional models of system reliability define failure in terms of boundary conditions, such as reaching unsafe states or executing inadmissible actions. Under these frameworks, systems are considered reliable as long as they remain within defined constraints. This work demonstrates that such evaluations are incomplete. A system may remain formally safe and admissible while progressively losing its ability to adapt, reconfigure, or recover within its own state space. As viable transition pathways collapse, the system can enter a functionally failed condition prior to any observable violation. To address this gap, the paper defines “transformability” as an independent system property capturing whether a system retains accessible pathways for state transition under its governing constraints. Loss of transformability represents a structural degradation of the state space, where alternative trajectories become limited or unavailable despite continued compliance with safety and admissibility conditions. The introduction of transformability establishes a third axis of system evaluation and reframes reliability as a continuous structural property rather than a binary condition tied solely to boundary crossing. This perspective has implications for AI systems, multi-agent reasoning architectures, and enterprise governance frameworks, where systems may appear stable and compliant while already lacking the capacity for meaningful progression or recovery. By identifying and formalizing failure before violation, this work provides a basis for earlier detection of system degradation and highlights the need for design and evaluation approaches that preserve structural flexibility over time.
Misty Michele Richards (Tue,) studied this question.